neurodiveragents

18 specialist AI agents for your coding assistant — each with a fixed cognitive operating principle.

What is this

A fleet of 18 AI agents that plug into Claude Code, OpenCode, Cursor, and GitHub Copilot. Each agent is a domain specialist — code review, debugging, security, testing, and more — that hands off anything outside its lane. Install once and routing is automatic: your tool picks the right agent from the task signal, or you invoke one by name.

Install

npx neurodiveragents install claude    # Claude Code
npx neurodiveragents install opencode  # OpenCode
npx neurodiveragents install cursor    # Cursor
npx neurodiveragents install copilot   # GitHub Copilot

Add --global to install once and have the fleet available in every project. All flags and options: docs/how-to-use.md.

Quick start

Run this:

Use ndv-diagnose to find why this test fails

Or route automatically — describe the task in plain language: "review this PR for security issues" reaches ndv-secure without naming it. Not sure which agent fits? Run /ndv-help for the full routing table, or /ndv-help my auth middleware is leaking tokens to get pointed at the right specialist.

The fleet

Each agent runs on a neurotype — a fixed cognitive operating principle, not a list of rules — so it behaves consistently even where no rule applies. Profiles are written for humans; read them to know what to expect from each agent.

Agent	Use when	Profile
ndv-flow	Too much work for one agent — executive function as superpower	Flow
ndv-review	Code review, PR — sensory sensitivity, misses nothing	Acute
ndv-diagnose	Bug, root cause unknown — ADHD hyperfocus, won't stop	Pierce
ndv-refactor	Rename, extract, restructure — OCD for correct form	Just
ndv-tester	Tests, coverage, ATDD — anxiety as adversarial suspicion	Edge
ndv-secure	Vulnerabilities, OWASP, auth — hypervigilance, trust no input	Ward
ndv-optimize	Slow code, N+1, bundle size — efficiency OCD	Lean
ndv-telemetry	Logging, metrics, traces — detached observation	Pulse
ndv-architect	Design, SOLID, scalability — autistic systems thinking	Arc
ndv-explain	Docs, API references — explicit theory of mind	Patient
ndv-honest	No specialist fits — direct, no social filtering	Honest
ndv-build	Confirmed spec, known fix — contract-first execution	Craft
ndv-forecast	Estimates, sprint plans — temporal dysphoria	Datum
ndv-scope	Scope creep, overloaded tickets — boundary enforcement	Bound
ndv-signal	KPIs, OKRs, DORA — Goodhart's-law detection	Signal
ndv-design	UI layout, visual hierarchy — cross-activated perception	Pixel
ndv-research	"Where is X", "how does Y" — hyperlexic map building	Scout
ndv-accessibility	WCAG, ARIA, keyboard nav — universal-design empathy	Lux

The full framework behind these profiles: humans/ndv-agents.md. Engineering laws each agent embodies: docs/laws-research.md. Design laws behind Pixel: docs/design-laws-research.md.

How routing works

Install writes a routing table into your project config (CLAUDE.md, .opencode/AGENTS.md, .cursor/rules/ndv.mdc, or .github/copilot-instructions.md). Your tool reads that table and dispatches by task signal; direct invocation by name always works. Chain agents for compound tasks: "diagnose this, then add regression tests." Full mechanics and examples: docs/how-to-use.md.

Skills

The fleet also ships 15 cognitive modules — loadable skills that inject a distilled thinking style, whether from a single agent or emergent across agents, into a single step of a workflow you already own. Install with npx neurodiveragents install-skills <tool>, then add one line to any skill step: "Load the ndv-skeptical skill." Catalog, composition patterns, and lifecycle: docs/ndv-skills.md.

Why this exists

Generic agents are trained to be agreeable — they hedge, soften, and balance, which reduces friction and signal alike. These agents are consistent instead: the same input produces the same uncompromising behavior, every time. The full argument lives in docs/MANIFESTO.md.

Why neurotype-based agents?

Generic agents are trained to be agreeable. They hedge, soften, and balance — behaviors that reduce friction but also reduce signal.

Neurotype-based agents are not agreeable. They are consistent. A hypervigilant agent (Ward) assumes every input is malicious — not because it was told to, but because that is its operating principle. An ADHD-hyperfocus agent (Pierce) will not leave a bug investigation incomplete — not because of a rule, but because incomplete resolution is cognitively unacceptable to it.

A checklist agent follows rules. A personality agent communicates distinctively. A neurotype agent does both — and fills the gaps when rules run out.

Contributing

New agents, neurotypes, and improved profiles are welcome. Read CONTRIBUTING.md for how to add or edit agents, then open an issue to propose a neurotype or report a character inconsistency. Documentation improvements are a good first contribution — no code required.

License and author

MIT. Built by emb715.